Why an Evolutionary System Needs Memory

I used to think an evolutionary system needed two things:
variation and selection.
Something changes. The system observes the result. Better variants survive.
That sounds reasonable — until you try to build one.
While working on an evolutionary trading system, we ran into a more basic problem:
If the system cannot reliably remember what changed, why it changed, and what happened afterward, it cannot prove that it is improving.
It can only prove that it is changing.
That distinction matters.
A system can modify parameters, promote agents, remove weak strategies, or generate new variants. From the outside, all of that looks like evolution.
But later you may want to ask:
Why was this agent promoted?
What evidence justified the decision?
What was the previous state?
What happened after the change?
Was the new state actually better?
If those answers cannot be reconstructed, the system has activity, but not real evolutionary memory.
Memory is more than storage
More logs are not enough.
More metrics are not enough.
Useful evolutionary memory needs structure.
A meaningful change should leave behind something like:
previous state → evidence → decision → change → resulting state
That chain gives us provenance.
It tells us not only what the system is now, but how it got there.
This is similar to software development.
A source file tells you what the code looks like today.
Git tells you how it became that way.
An evolutionary system needs the same kind of lineage.
Sometimes evolution means doing nothing
There is another consequence.
An evolutionary system should not change simply because it has permission to change.
If no candidate satisfies the evidence threshold, the correct action may be:
no change.
That is not failure.
It is discipline.
A system that always changes produces novelty.
A system that remembers, evaluates, and sometimes refuses to change has a chance to produce learning.
So my original equation:
Evolution = variation + selection
has changed.
I would now write:
Evolution = variation + selection + verifiable memory
Because without memory, selection cannot accumulate knowledge.
And without accumulated knowledge, a system may keep changing while repeatedly rediscovering the same mistakes.
The better question is no longer:
Can the system change itself?
It is:
Can the system explain, with evidence, why its current state exists?
If it can, then we may be getting closer to a system that actually evolves.